Agent-native infrastructure for production AI
Define agents as code. Deploy with a single command. Astro handles models, knowledge bases, tool integrations, and observability so you build what matters.
Everything agents need to run
Purpose-built infrastructure primitives, not repurposed web hosting.
Models
Attach any LLM (OpenAI, Anthropic, open-source) declaratively. Swap providers without touching code.
Knowledge Bases
Vector-indexed knowledge stores with automatic ingestion and chunking. Your agents stay grounded.
Tool Integrations
Connect APIs, databases, and SaaS tools. Agents call them natively with built-in auth and retries.
Observability
Traces, metrics, and logs out of the box. See every decision your agent makes in production.
Guardrails & Secure Isolation
Each agent runs in its own container with scoped credentials. Zero shared state between deployments.
Auto-scaling
From zero to thousands of concurrent sessions. Infrastructure that responds to demand, not configuration.
Declarative by design
The Astro spec (astropods.yml) is your agent's single source of truth — models, knowledge stores, tools, and observability — all defined in one file.
name: support-agent
version: 1.0.0
container:
image: astro/support-agent:latest
runtime: python3.11
models:
- provider: openai
model: gpt-4o
role: reasoning
knowledge:
- source: s3://docs-bucket
index: vector
embeddings: text-embedding-3-small
tools:
- name: ticket-lookup
type: api
endpoint: https://api.internal/tickets
- name: slack-notify
type: integration
channel: "#escalations"
observability:
tracing: true
metrics: trueThree commands to production
Shared knowledge sources and consistent deployments mean you experiment and ship to production faster.
Scaffold
$ ast create my-agentOne command generates the project structure and a starter Astro spec so your agent is fully specified as repeatable infrastructure.
Develop locally
Run your agent locally with hot-reload. Test against live models and tools with built-in adapters for voice, text, Slack, and streaming interfaces.
Deploy
Push to the Astro registry. Your agent is live with tracing, scaling, and versioning, available to others in your organization to use.
Built different
Not just another hosted runtime. Astro gives you control, flexibility, and independence at the infrastructure level.
Live updates, zero downtime
Change agent configurations, swap models, update guardrails, and adjust policies in production without redeploying. Your agents keep running while you iterate.
Framework-agnostic
Bring agents built with CrewAI, LangChain, Mastra, or your own custom framework. Astro doesn't lock you into a single SDK. It runs whatever you build.
Self-host or managed
Run Astro in your own cloud, on-prem, or use our managed platform. Your agents, your data, your infrastructure — with no vendor lock-in.
An Open Ecosystem
Build your agents using popular frameworks of your choice. Frameworks are adapted with a unified SDK to make using any framework as easy as a single import.
Built for real workloads
Real agents built and deployed by teams on Astro, complete with ingestion and inference engines. Because you can't deploy real agents without first building the right knowledge sources.
Searches and summarizes documentation across multiple sources with context-aware retrieval.
Reviews pull requests for bugs, security issues, and style violations with full repo context.
Generates and A/B tests email campaigns, landing page copy, and social content.
Triages and resolves support tickets end-to-end with knowledge base lookup and escalation.
Scores and qualifies inbound leads using CRM data, enrichment APIs, and custom criteria.
Monitors application logs in real-time, detects anomalies, and surfaces actionable insights.
Make your agent
your own
Every agent gets a verifiable identity. Track provenance, capabilities, and performance at a glance.
- Verifiable provenance and ownership
- Real-time performance metrics
- Capability tagging and tier classification
- Tamper-proof registry records
Open source
to the core
Astro is open source and Apache 2.0 licensed. We believe the best agent infrastructure should be transparent, extensible, and community-driven.
Contribute & earn free compute
Build and share agents on the platform. Contributors earn free compute credits to run their agents in production.
Share with your team or the world
Deploy agents privately within your organization or publish them to the community for anyone to use and customize.
Join the community
Be part of a growing community of agent developers building, iterating, and shipping real AI agents together.
Start building agents
that work for you
Free to start. No credit card required. Deploy your first agent in minutes.